Machine learning risk stratification to identify people living with HIV at high risk of delayed ART and advanced immunosuppression: a precision public health approach

Description

Jinling Yin and colleagues published a 2026 study building an AI tool to flag people living with HIV at risk of starting treatment late. Working with de-identified records for 5,436 people diagnosed from 2016 onward, they defined high risk as delayed antiretroviral therapy or a CD4 cell count under 200 at baseline. That state marks advanced immunosuppression. Six standard classifiers were trained and then combined into a stacking ensemble, a model learning from the others' outputs.

The high-risk label applied to 61.2 percent of the cohort, and the ensemble identified them with an AUC of 0.915, holding precision at 0.810 even when tuned to catch at least 90 percent of cases. Baseline CD4 count, age, and transmission route were the strongest predictors. Decision analysis showed the tool beat blanket strategies of intervening on everyone or no one. The authors frame it as precision public health, while insisting on prospective testing and fairness checks first.

Catalog record

Source: https://europepmc.org/articles/PMC13279522